Metacognitive Networks and Measures of Consciousness

Abstract

Subjective measures of awareness rest on the assumption that conscious knowledge is knowledge that participants know they possess. Post-decision wagering, recently proposed as an objective measure of awareness, raised a new controversy on determining the properties that should characterize the objectivity of an awareness measure. Indeed, if the method appears objective in many aspects – it does not require introspection but rather lies on instinct, it does not affect conscious states, it can be learned unconsciously –, it also shares some characteristics with subjective measures – it involves metacognitive content and particularly, it represents a decision about a decision. The lack of consensus on this topic leaded us to develop a new approach based on a novel theoretical aspect, causality, and to consider a causally independent mechanism that would give an agent the capability to know what knowledge it possesses. In this framework, any measure that would not necessarily rely on such mechanism in a given experimental situation should be considered as objective. We support our claim with a computational model based on metacognitive networks, and present three simulation studies in which neural networks learn to wager on their own performance. Results demonstrate a good fit to human data, although depending on the situation, post-decision wagering is implemented either as an objective or as a subjective measure of network’s knowledge. We discuss implications of our results for defining the nature of subjective and objective measures, as well as for our understanding of consciousness.


Back to Saturday Papers